Towards sustainable concrete: Estimating the compressive strength of SCM-enriched concrete via machine learning techniques

The growing demand for sustainable and high-performance concrete has accelerated the adoption of supplementary cementitious materials (SCMs) as partial replacements for ordinary Portland cement (OPC). This study proposes a comprehensive machine learning (ML) framework for predicting the compressive strength of SCM-blended concrete using seven regression algorithms, namely XGBoost, Gradient Boosting, CatBoost, Random Forest, LightGBM, Support Vector Regression (SVR), and k-Nearest Neighbors (KNN). A dataset comprising 1456 experimental concrete mixtures was employed, incorporating mixture design variables including cement content, binder content, water content, aggregate content, water-to-binder ratio, and OPC replacement level. Model hyper parameters were optimized through randomized search, and predictive performance was evaluated using 10-fold cross-validation. Among the evaluated models, XGBoost achieved the best overall performance, yielding an average coefficient of determination (R² = 0.8016 ± 0.0433), MAE = 5.19 ± 0.51 MPa, and RMSE = 7.35 ± 0.76 MPa, while also exhibiting the highest correlation coefficient (R = 0.900). To enhance model interpretability, SHapley Additive exPlanations (SHAP) analysis identified the water-to-binder ratio, cement content, and binder content as the most influential variables governing compressive strength prediction, with dependence plots revealing distinct non-linear relationships between these predictors and the model output. An offline graphical user interface (GUI) prototype was developed in Python based on the optimized model to demonstrate the rapid estimation of compressive strength from concrete mixture parameters. The proposed framework provides an interpretable ML-based approach for supporting the evaluation and design of sustainable SCM-blended concrete mixtures, while the GUI prototype demonstrates its potential for practical application.